Instrument transformers CT / VT

What it is (precise)

A current transformer is connected in series with the circuit it measures. Its primary is the power conductor itself, usually a single pass so the primary turns count is one, and its secondary delivers a scaled, galvanically isolated replica of the primary current into a low-impedance burden. Seen from the CT's own terminals the primary behaves as a current source — the primary current is set by the power system, whose source impedance is enormous next to the impedance the CT reflects back into it — so the CT has no influence on the quantity it is measuring.

The device is ampere-turn balance and little else: N(primary) × I(primary) = N(secondary) × I(secondary) + N(secondary) × I(exciting), which makes I(secondary) ≈ [N(primary)/N(secondary)] × I(primary) essentially independent of what the secondary is wired into, until the core saturates. Burden behaviour, saturation and the open-circuit hazard are all consequences of that current-forced character.

A voltage transformer is connected in shunt across the primary system and steps system voltage down to a 110 V or 120 V class secondary. It behaves as a voltage source behind a small series impedance — V(secondary) = V(primary)/K − I(secondary) × (R + jX) of the winding, with I(secondary) = V(secondary)/Z(burden) — so its output is set by the primary voltage and the turns ratio, and is degraded, never raised, by burden current. It is the exact circuit dual of a CT, and most of the practical rules below are one dual statement read in both directions.

The device also carries two names: potential transformer (PT) is the older North American term and voltage transformer (VT) the IEC-derived one, and they denote the same shunt-connected device — this site's one-line-diagram entry writes PT where the revenue-meter and protection-relay entries write VT. A drawing that carries both is using two names for one device, and the fix is a note on the legend rather than a change of hardware.

Four CT constructions turn up on a BESS one-line. A window (ring) CT has no primary of its own: the cable or busbar passing through the aperture is the primary, one turn. A bar CT is the same core with that single-turn primary built in as a solid bar. A bushing CT is a window core mounted around the insulating bushing of a transformer or breaker, which is why main-transformer and breaker CTs are commonly bushing units — the housing is already there.

A wound CT carries a multi-turn primary winding, used where the primary current is small enough that one turn cannot produce the ampere-turns the core needs, which is ampere-turn balance again. On the voltage side, an inductive VT is a small wound transformer connected across the primary, while a capacitive voltage transformer (CVT) taps a capacitive divider on the line and feeds a smaller inductive unit from that tap. A CVT's accuracy and transient behaviour are stated in its own specification, so read them rather than assuming an inductive unit's figures carry across.

Burden, and which way it moves the error

Burden is what the secondary has to drive outside itself: the connected relay and meter inputs, the two-way lead resistance, and test-switch and terminal contacts. It is quoted either in ohms or as VA at rated secondary current, and the two forms convert through VA = I(secondary)² × Z, so a VA figure means nothing until the secondary current it is stated at is named. Ratings are defined at the secondary terminals, so the CT's own winding resistance sits outside the burden and is added separately for a saturation check: E(secondary) = I(secondary) × [R(ct) + R(leads) + Z(devices)].

Modern numerical relay and meter current inputs present a fraction of a VA — order 0.02 to 0.1 VA is typical — against the several VA of the electromechanical relays the classical sizing rules were written around, so on a large BESS site the connected burden is dominated by lead resistance rather than by the number of devices. Route length and conductor cross-section are the sizing variables that decide it.

One detail decides whether the leads are counted once or twice. For a star-connected CT group sharing a residual return conductor, a balanced three-phase fault gives I(a) + I(b) + I(c) = 0, the residual conductor carries nothing, and each phase loop contains one lead length.

A single-phase-to-earth fault sends current out on the phase lead and back on the residual lead, so the loop contains two. Worked: 150 m of 4 mm² copper is about 0.68 Ω one way — 17 VA at 5 A on the phase fault, 34 VA on the earth fault, already past a 30 VA core. The earth-fault loop is the higher-impedance one, and using the single-length figure understates its lead burden by a factor of two.

Which case sizes the core is a different question, because required EMF is fault current times loop impedance, and the lead term is only part of that loop — the winding resistance and the devices sit in it too. Whether it governs depends on how the two fault currents compare, so run both cases rather than assuming either.

On a low-resistance-earthed collector, or on a delta winding where no earth-fault current reaches the phase CTs at all, the three-phase case governs instead — the grounding-system entry owns why those currents differ by orders of magnitude. The split applies to a star-connected group with a common residual return, and a dedicated two-wire run always counts both conductors.

Raising the burden makes a CT's ratio and phase errors worse and never better, and it moves the ratio error in the negative direction.

The chain is short: more burden raises the required E(secondary), higher E(secondary) means higher peak flux through E(secondary) = 4.44 × f × N(secondary) × A × B(max), higher flux drives the core further up the non-linear magnetisation curve and raises the exciting current disproportionately, and since I(primary)/N = I(secondary) + I(exciting), every ampere diverted into the exciting branch is an ampere missing from the output. Relative to its actual turns ratio an uncompensated CT therefore reads low.

Real metering cores are turns-compensated — a fraction of a turn or one whole turn removed from the secondary — so the residual error straddles zero across the class range, which is why accuracy classes and ratio correction factor limits are two-sided; take the monotonic direction as absolute and treat the residual error itself as a two-sided band.

The sign of the phase displacement is set by the burden's power factor rather than by the CT: with the resistive burden a numerical relay or meter presents, the secondary current leads (positive displacement by the IEC convention), and with a strongly inductive burden the sign reverses, while the magnitude error stays negative either way.

Magnitudes come from the class rather than from a rule of thumb — class 0.2 permits ±10 arcminutes (0.167°) at rated current, 5P permits ±60 arcminutes (1°, 1.8 crad). A loaded VT reads low by the dual argument, V(secondary) = [V(primary)/K] × Z(b)/[Z(b) + Z(eq)], which sits below the value its own turns ratio would give for any finite burden and falls monotonically as burden VA rises, the reactive part of the drop rotating the output so it lags.

The absolute sign is a separate matter: metering VTs are compensated by adding secondary turns, so a compensated unit is corrected to centre its error across the specified burden range and can read high at light burden — hence VT classes specified over a burden range, typically 25 to 100% of rated, rather than at one point.

Both magnitude errors sit in the same direction and compound across the chain; the revenue-meter entry owns what that does to a settled number, and the p-q-capability-test entry owns why the angular part lands harder on reactive power than on real power near unity power factor.

Standard secondaries, and how the ratios are written

Standard CT secondaries are 5 A (ANSI and North American practice, short leads) and 1 A (IEC practice, preferred for long runs). The reason is the square in VA = I(secondary)² × Z. For the same cable run a 5 A CT sees 25 times the burden VA of a 1 A CT, and for the same VA rating a 1 A CT tolerates 25 times the ohms. Worked: a 100 m route — 200 m of loop — in 2.5 mm² copper is about 1.4 Ω, which loads a 5 A CT with 35 VA and a 1 A CT with 1.4 VA.

Expressed in volts rather than VA the same leads demand one-fifth of the secondary EMF at 1 A, and the two framings agree: a 1 A core of the same primary rating carries about five times the secondary turns, so its achievable knee-point voltage is also about five times higher for the same iron, and the net core-duty advantage returns to 25 times.

On a site where the switchyard and the control building are a few hundred metres apart, 1 A secondaries are the choice that survives the cable schedule. The higher turns count has one cost worth knowing: it also means a higher R(ct) and a higher open-circuit voltage should the secondary ever be broken.

Standard VT secondaries are 120 V (ANSI) and 110 V (IEC), with 115 V and 100 V also in service, plus the line-to-neutral forms 110/√3 = 63.5 V and 120/√3 = 69.3 V. Ratios are written primary:secondary at rated conditions — 600:5, which is 120:1, or 34500:120, which is 287.5:1.

A wye-connected VT set on a 34.5 kV system is marked 34500/√3 : 120/√3, or 19919:69.3, and that is the same 287.5:1 ratio: the √3 form states the rated primary voltage applied across each unit line-to-neutral, a separate quantity from its rated insulation level, and three such units in wye still present 120 V line-to-line at the secondary.

What injects a √3 is crossing the two forms — pairing the line-to-line primary number with the line-to-neutral secondary number, 34500:69.3 = 498 or 19919:120 = 166, both wrong by √3 — or entering the wrong secondary nominal base or VT-connection setting in the relay.

A different error is physical: a 19919 V unit applied line-to-line is over-excited by 73% while its secondary still reads a believable 120 V. A broken-delta (residual) winding is a separate winding from the wye measuring winding, rated so that three windings in series produce full nominal secondary voltage for a complete zero-sequence displacement — 110/3 = 36.7 V or 120/3 = 40 V per winding, giving 110 V or 120 V of residual output; other residual-winding ratings are used depending on the system earthing arrangement, so read the nameplate rather than assuming this one.

Measuring residual voltage at all requires a wye-connected set with the primary star point earthed, because line-to-line voltages contain no zero-sequence component and a delta set, or a wye set with an unearthed star point, cannot reproduce it. The current-side equivalent — residual sum versus core-balance CT, and why both see 3 × I(0) — belongs to the zero-sequence and protection-relay entries.

The arithmetic direction is worth stating explicitly because inverting it is expensive. With K(n) = I(primary)/I(secondary), a primary quantity is the secondary quantity multiplied by K(n), and a relay setting expressed in primary amps is divided by K(n) to become a secondary-amp setting. On a 600:5 CT, K(n) = 120: a 4.2 A secondary reading is 504 A primary, and a 1,000 A primary pickup is 8.333 A secondary.

Inverting the operation misplaces the setting by K(n)² — 14,400 times here — and nothing on the panel objects, because the element simply does not pick up until a fault goes uncleared. This is why a settings sheet carries every threshold twice, once in primary units from the fault study and once in secondary units for injection at commissioning; the protection-relay entry owns that sheet and who signs it.

The rule that has to be right: never open a live CT secondary

An energised CT secondary must never be open-circuited, and short-circuited is the safe state. Apply the shorting link, shorting screw or make-before-break test switch before breaking the loop, and put no fuse or other series protective device anywhere in a CT measuring circuit. A VT is the mirror: its secondary must never be short-circuited, open is the safe and most accurate state, and it is protected by fuses or an MCB.

Both halves belong together, because a technician who learns only one of them will generalise the wrong way onto the other device. Memorise it by source type — a current source wants its loop closed, a voltage source wants its loop open and loaded rather than shorted. The same rule covers spare capacity: an unused protection core is shorted at the terminal block, never left open.

The mechanism explains the severity. In normal operation the net magnetising ampere-turns are a small residue, typically well under 1% of the primary ampere-turns, and the core sits low on its magnetisation curve. Force the secondary current to zero and ampere-turn balance becomes N(primary) × I(primary) = N(secondary) × I(exciting): the entire primary ampere-turn product becomes magnetising ampere-turns, two to three orders of magnitude above design.

The primary current cannot fall in response, because the power system imposes it. The core is driven hard into saturation on each half cycle, the flux waveform becomes a clipped near-square wave, and the induced secondary voltage — N(secondary) × dΦ/dt — is near zero across the flat saturated portions and impulsive during the fast flux transitions near each primary-current zero crossing. The result is a train of narrow peaks reaching kilovolt level, as peaks rather than as an RMS figure.

The consequences are a lethal shock hazard, insulation failure, core overheating and permanent remanent flux left in the iron. The hazard grows with primary current and with turns ratio, so a fully loaded MV feeder CT at a high ratio is the worst case rather than the benign one, and the moments of exposure are ordinary maintenance moments: swapping a meter, lifting a lead at a test block, or breaking a loop to inject current while the feeder is still carrying load.

Two habits close most of that exposure. First, treat the shorting hardware as part of the work method — apply it, verify it, then open the circuit. Second, remember that a DC winding-resistance measurement magnetises the core the same way a fault does and must always be followed by demagnetisation, or the next event starts from a core that is already part-way to saturation.

Sizing a protection core past the class label

A class label states what the core does at rated burden; what it does in your circuit takes three more numbers. The first is the knee point. The IEC knee-point EMF is the secondary EMF at which a 10% increase in voltage produces a 50% increase in exciting current — the bend in the magnetisation curve as the flux density approaches saturation — and the design condition is that the worst-case required EMF stays under it: I(secondary) × [R(ct) + R(leads) + Z(devices)] × K(td) < V(k).

IEEE C57.13 defines the knee by a tangent construction on log-log excitation axes instead, so IEC and ANSI knee-point figures are not interchangeable and any V(k) on a datasheet needs the standard named beside it.

The excitation curve measured at the secondary terminals is the record of all this and is worth having in the test file. The second number is the accuracy limit factor, which is burden-dependent: the printed figure applies at rated burden, and the effective limit rises at a lower actual burden and falls at a higher one, roughly as ALF(effective) ≈ ALF(rated) × [R(ct) + Z(b, rated)]/[R(ct) + Z(b, actual)].

A core rated at 20 times rated current with R(ct) = 2 Ω and a 5 Ω rated burden, wired into 1 Ω actual, gives about 47 times — so reducing lead burden buys fault-current headroom without changing the CT. The relation holds in the linear region only, the real core's knee sets a ceiling on it, and omitting R(ct) from the arithmetic overstates the improvement.

The third number is what actually drives the core into saturation, and there are four contributors. Flux follows the voltage integral, so fault current magnitude and burden enter as a direct product — doubling either doubles the flux demand — while DC offset and remanent flux act on top. DC offset dominates.

A fully offset fault current contributes a unidirectional flux component whose ratio to the AC component is X/R, so the worst-case demand is (1 + X/R) times the symmetrical-current value: at X/R = 10 the core needs 11 times the flux, which is usually uneconomic to design out. That is why the practical requirement is normally written as a time to saturation longer than the relay's decision window rather than as no saturation at all, and why a stiff close-in source is the hard case even when the fault magnitude alone looks manageable.

Remanence is the fourth: after a fault is interrupted, a gapless core is left sitting on its hysteresis loop with no mechanism to reset, and can retain up to roughly 80% of saturation flux. If the next event drives flux the same way, only about 20% of the swing remains and time to saturation falls by roughly a factor of five — the autoreclose-onto-a-persistent-fault case — while the opposite polarity performs better than nominal.

The sign is effectively random, so design assumes the adverse case. Anti-remanence gapped cores (IEC class TPY, or class PR with a specified remanence factor) hold remanence to about 10%. Note also that a class P guarantee covers steady-state symmetrical current only and says nothing about DC offset or remanence; transient duty is covered by the TPX, TPY and TPZ classes, or by an explicit transient dimensioning factor K(td) applied to the required secondary EMF.

Saturation fails in two opposite directions, which is why matching matters as much as absolute capability. Once the core saturates, the secondary output collapses over most of the cycle and appears only in brief windows near the current zero crossings, and the RMS and fundamental content of that clipped waveform are lower than the ideal scaled value.

Every consequence sits in the fail-to-trip or slow-to-trip direction: an instantaneous element set at 8 times rated may see an apparent 6 times and hold; an inverse-time element evaluated at a lower multiple takes proportionally longer on its curve; a distance element computing impedance from a current that is too small computes an impedance that is too large and therefore under-reaches, seeing the fault as farther away and pushing a zone-1 fault into zone-2 timing. The differential case runs the other way.

On an external through fault, ideal CTs at the two ends cancel, but if one end saturates and the other does not, the differential quantity becomes a large fraction of the through-fault current and can cross the restraint characteristic, tripping a healthy transformer, bus or BESS block. Unequal burdens, unequal remanence, different cores or ratios and different DC offset at the two ends all produce it, and the cures conflict — raising the percentage-restraint slope stops the false trip and costs internal-fault sensitivity.

For a differential scheme, match the CTs' burdens and knee points to each other, not only to a specification. The opposite requirement on the metering side — a core specified to fold early so fault current stays out of the meter — is the revenue-meter entry's subject; the one point worth adding here is that the instrument security factor is defined at rated burden too, so an under-burdened metering core stays linear longer and therefore protects its meter less than the printed figure implies.

Cores, secondary circuits and the walkdown

Metering and protection cores at the same location are separate cores sharing one primary bar in one housing — a multi-core CT — because the two jobs want opposite things from the same iron. Allocation then follows the job. The main revenue meter takes the metering core with the tightest burden control, at the metering class the offtake agreement names. The check meter takes a separate core or winding, so a failure shows up as disagreement between two records rather than as a settlement query months later.

The feeder or transformer relay takes a protection core sized against the fault duty, including its X/R. The plant controller's transducer is a control measurement affecting regulation quality rather than settlement, so it can share a core provided its burden is counted in the circuit total. That last clause is the practical constraint on the whole allocation: every device added to a secondary circuit adds burden, so adding one to an existing revenue CT circuit is a contractual change rather than merely a wiring change.

Across a grid-scale plant the devices show up in four places: revenue CTs and VTs at the interconnection; protection CTs on the MV breakers in the collector switchgear and around the main transformer, feeding 50/51, 87T and 87B; CTs and VTs at each MV/LV inverter transformer for feeder protection, with the PCS's own LV current sensing acting as a control device rather than a revenue one; and auxiliary CTs and VTs on the station-service transformer, plus controller feedback for AGC and voltage regulation.

Where the revenue set sits relative to the contractual boundary changes the settled energy: the point of interconnection is usually the high-voltage side of the main step-up transformer while the metering set sits on the medium-voltage side, so on export the meter reads before the transformer losses and the compensation subtracts them; on import it adds, the loss falling against the plant either way, and the revenue-meter entry owns how that is computed.

Three relay-facing consequences are owned elsewhere and belong on the cross-reference list rather than in a restatement here: a CT-fed transformer differential across a Dy or Yd unit must apply the clock-number × 30° correction and remove zero sequence, or a standing differential appears on normal load, which the transformer-vector-group entry owns; no zero-sequence current reaches CTs on the delta side of such a transformer for a wye-side ground fault, which is a measurement blind spot as well as a circuit one, covered in the zero-sequence entry; and a VT or CT pair landed against the wrong phase shifts every product by 120° while every magnitude on every nameplate stays correct, which the phase-sequence entry owns.

Polarity is the check that catches the rest: marks are P1/P2 on the primary and S1/S2 on the secondary, current entering P1 leaves S1, and the marks are verified physically rather than assumed from the drawing.

The secondary circuits have their own rules. Earth each instrument-transformer secondary circuit at one point only, and let the schematic be the record of where that point is, because a second earth puts a second path around the measuring loop and no instrument reports it. VT secondaries are protected by fuses or an MCB — and CT secondaries are protected by nothing in series, as above — so a blown VT fuse is a live failure mode on a metering installation, which the revenue-meter entry describes from the meter's side.

Wound VTs on systems without a solid earth reference also raise ferroresonance, an interaction between the non-linear magnetising branch and system capacitance; it is a system-study and damping question rather than a datasheet one, so ask which study covered it and what the VT specification says about damping before ordering.

On standards, cite the part rather than the series: current transformers sit in IEC 61869-2 and inductive voltage transformers in IEC 61869-3, so a specification that names 61869 bare has named a series, not a requirement. Finally, MV switchgear increasingly ships low-power instrument transformers and Rogowski-type current sensors, whose secondary is a low-power signal into an electronic input rather than a 1 A or 5 A circuit.

The burden arithmetic and the shorting hardware on this page are written for iron-cored units with conventional secondaries, so treat a low-power sensor as a different interface: check the relay or meter input it is specified against, and the sensor manufacturer's own cable and termination rules.